TWO-LEGGED robots like Sony’s diminutive Qrio and Honda’s child-size Asimo might give the impression that they’ve got the problems of locomotion licked. But the robots are always put through their paces in carefully controlled environments with surfaces on which their creators know the bots won’t slip.
This is because moving from one type of surface to another that gives a different friction can make these robots unsteady on their feet. It is possible to use cameras and an image-processing computer to give robots a sense of the surface they are standing on, as well as the next one they will encounter as they walk or trundle along. But this is an expensive option – and one that adds power-draining weight to the robot.
But Penny Probert Smith and Konstantinos Zografos at the University of Oxford may have a disarmingly simple answer to this problem: sonar. Writing in the journal Robotics and Autonomous Systems (vol 51, p 17), the pair suggest adapting sonar systems, which many robots already use to avoid obstacles, to the task of recognising surface textures.
Advertisement
Sonar uses a loudspeaker to transmit a powerful sound wave that bounces off objects it strikes. The system works out where the objects are by measuring the time it takes for the echo of the sound wave to bounce back to the receiver. But the sequence of sound intensity values and frequencies that make up the echo also contains telltale information on the reflecting properties and granularity of the surface it bounced off.
The Oxford team used an ultrasound sonar on a small wheeled robot to build up a profile of common surfaces. They constructed sonar surface “signatures” for carpet, asphalt, plastic, gravel, hedges, flower beds, short grass and long grass. After averaging the signatures to reduce noise effects, the sonar algorithm guessed the surface correctly in repeated tests, the pair report. “And [unlike a vision-based approach] it works in the dark,” says Probert Smith.
The technique should mean that robots can better plan routes over “friendly” surfaces and work out in advance how to achieve the correct traction so they don’t slip. Walking robots could do this by changing their gait, for instance, while other bots could turn their wheels more slowly so they don’t lose their grip.
